Premium
WE‐F‐105‐01: On the Importance of Nuclear Models On the Accuracy of Fast Monte Carlo Methods for Proton‐Therapy
Author(s) -
Souris K,
Lee J,
Sterpin E
Publication year - 2013
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4815617
Subject(s) - imaging phantom , monte carlo method , physics , proton , stopping power , proton therapy , neutron , computational physics , nuclear physics , nuclear medicine , mathematics , optics , detector , statistics , medicine
Purpose: Parallel architectures like graphical processor units (GPU) with high computational power allow dose distributions for proton beams to be calculated with Monte Carlo (MC) simulations within the time constraints of clinical practice. However, physical models often need to be simplified to exploit the full potential of GPUs. This study aims at evaluating comprehensively the impact on computation accuracy of the simplified model for nuclear inelastic collisions described by Fippel et al (Med Phys 2004) and used by Jia et al (PMB 2012). Methods: Three MC codes were used: 1) an in‐house version of Fippel's algorithm, which uses an empirical method to reproduce approximately ICRU 63 nuclear inelastic cross‐sections for oxygen; 2) an extension of PENELOPE to protons (PENH) that samples directly ICRU 63 cross sections (but neglects neutrons); 3) GEANT4 with Binary‐Cascade.The integral depth‐dose deposited by a 200 MeV proton beam into water was then calculated using those MC codes. The simulations were performed in water phantoms of dimensions 4.1×4.1×40, 8×8×40 and 60×60×40 cm 3 with a scoring resolution of 1 mm in depth. Results: In the 4.1×4.1×40 cm 3 phantom, deviations between Fippel's algorithm and GEANT4 were within 2%, whereas this bound grew up to 3% between PENH and GEANT4. However, when the phantom dimensions are increased, the deviations between PENH and GEANT4 remained within 3%, while they reached 5% for the 60×60×40 cm 3 phantom using Fippel's method. Conclusion: Our in‐house implementation of Fippel's algorithm did not reproduce consistently GEANT4 results for varying phantom sizes. This was attributed to the over‐simplified sampling of the angular distributions of secondary protons resulting from nuclear inelastic reactions. However, this was not observed for PENH that implements an exact sampling of ICRU 63 cross‐sections. Therefore, the details of nuclear models may have a significant impact on accuracy. Kevin Souris is financed by the Walloon Region under the project name InVivoIGT, convention number 1017266. Kevin Souris is sponsored by a public‐private partnership IBA ‐ Walloon Region